Harnessing the Power of Large Language Models: Automatically Generating High-Quality, Readable, and Maintainable Unit Test Cases for Java and Python

In this talk, Lisa Waugh will discuss the challenges in delivering high-quality software due to manual testing and the need for automated unit test generation. Despite the development of techniques and tools to automatically generate unit tests, the generated tests often suffer from poor readability and lack resemblance to developer-written tests. To address these issues, Lisa's team has investigated the use of large language models (LLMs) to improve the readability and usability of automatically generated tests. They have created a pipeline to guide LLMs in generating high-coverage, usable test cases for Java and Python, with plans to expand to other languages. The unit test cases include mocking so they can be run in a CI/CD pipeline prior to deployment. Lisa will present the results of their LLM-based test generation with statistical analysis guidance and compare the results of that work with existing test generation methods. She will also introduce a product that enables developers to create readable and runnable tests with good coverage.